Re: [RFC PATCH] gpio: support for GPIO forwarding

[Alexandre Courbot <gnurou@xxxxxxxxx>]

On Wed, Jan 14, 2015 at 9:58 PM, Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
> On Thu, Dec 18, 2014 at 9:23 AM, Heikki Krogerus
> <heikki.krogerus@xxxxxxxxxxxxxxx> wrote:
>
>> This makes it possible to assign GPIOs at runtime. The
>> motivation for it is because of need to forward GPIOs from
>> one device to an other. That feature may be useful for
>> example with some mfd devices, but initially it is needed
>> because on some Intel Braswell based ACPI platforms the GPIO
>> resources controlling signals of the USB PHY are given to
>> the controller device object for whatever reason, so the
>> driver of that controller needs be able to pass them to the
>> PHY device somehow.
>>
>> So basically this is meant to allow patching of bad (bad
>> from Linux kernels point of view) hardware descriptions
>> provided by system fw in the drivers.
>>
>> Signed-off-by: Heikki Krogerus <heikki.krogerus@xxxxxxxxxxxxxxx>
>> ---
>>
>> Hi,
>>
>> I'm sending this first as a RFC in case you guys have some better
>> idea how to solve our problem. I actually already have couple
>> platforms where the GPIO resources are given to the "wrong" device
>> objects now.
>>
>> Originally we were thinking about simply handling our problem with
>> hacks to the PHY drivers, basically also checking if the parent has
>> GPIOs. That would only work if the controller is always the parent,
>> which it's not, but even if it was, it would be too risky. The PHY
>> drivers don't know which controller they are attached to, so what is
>> to say that the GPIOs aren't really attached to the controller.
>>
>> So the safest way to handle our problem is to deal with it in quirks
>> in the controller drivers. Solving it by bouncing the GPIOs did not
>> feel ideal there doesn't seem to be any other way. The API is copied
>> from clkdev (clk_register_clkdev). In the end it's really only one
>> function for adding the lookups and one for removing them.
>>
>> The way it's implemented is by modifying the current style of handling
>> the lookups a little bit. The per device lookup table is basically
>> reverted back to the single linked-list format since it seems that
>> these lookups may have to be assigned from several sources.
>
> Oh ain't that great.
>
> So now we start patching around the kernel because the ACPI
> tables are erroneous for GPIOs. I'd like some feedback from
> Rafael &| Darren on this patch, i.e. if you two think this is a good
> way of accounting for bad GPIO descriptions in ACPI tables then
> ACK this patch.
>
> I guess the other option would be to fix up the ACPI DSDT
> itself to put resources right, correct? Is this not possible?
>
> Alexandre also need to ACK it before I dare do anything with
> it.

I am not really fond of this idea since it adds complexity to the
(already too complex) GPIO lookup, and only solves to a local level
(GPIO) what is a more global problem (bad ACPI tables that can affect
any subsystem).

Also I don't think any new functions is actually needed: on an ACPI
system we can safely assume that the platform lookup tables are not
used at all. So, although it is a hack as well, you can just call
gpiod_add_lookup_table() with a runtime-built table from the driver
that needs to pass the GPIO so the receipient can receive it through
the lookup table fallback gpiod_get() uses if no GPIO is defined via
ACPI.

So I think that even with the current state of the code you can
achieve what you need. Should you do it, that's another question - it
seems more to-the-point to find a way to fix/patch the ACPI tables at
runtime, if that is possible at all, to provide a more general
solution to this issue.
--
To unsubscribe from this list: send the line "unsubscribe linux-gpio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html